search

jiyanggao / TALL

TALL: Temporal Activity Localization via Language Query
188 stars 48 forks source link

About the C3D feature #1

Closed mengliu1991 closed 6 years ago

mengliu1991 commented 6 years ago

Hi,

I download the feature extraction file from the c3d GitHub web, but I find that they can only extract features for 16 frames. Can you provide your retrain C3D model ( 128 frames, 64 frames, 256 frames) for TALL dataset for me?

jiyanggao commented 6 years ago

Hi, I didn't retrain the C3D model. I just uniformly sample 16 frames from the 128 frames, and input them into the C3D.

h-bo commented 6 years ago

So you mean you feed them separately and mean the feats at last as feat of 128 frames?

h-bo commented 6 years ago

Can you give me feat for per 16 frames? Thanks

stha-prashant commented 2 years ago

hello @h-bo, I'm looking for the same answer, did you find the c3d model used for feature extraction?